Virtual environments present what appear to be conflicting demands on scenario control. On one hand, computer generated agents must behave in consistent and believable ways in a complex, dynamic environment. They must interact with other simulated agents and subjects who have considerable freedom of action. On the other hand, experimental and training applications require that subjects be tested under controlled conditions. The essential aspects of events and situations must be repeated from trial to trial. The challenge we face is to create scenarios that reproduce the intended conditions without overly restricting the subject's actions and while maintaining dynamism, complexity, and spontaneity in agent behaviors.
For example, consider the problem of creating a crash threat on a simulated urban freeway for a virtual driving environment. Such a scenario might be part of an experiment to determine the influence of Alzheimer's disease on the driving ability. The scenario requires generation of dense traffic that provides a backdrop for a critical situation such as an abrupt stop or dangerous lane change. In order to compare the performance of subjects, the behavior of the vehicles surrounding the driver must be carefully orchestrated -- some vehicle, possibly determined on-line, must perform the threatening deed and gaps in the traffic that provide possible escape routes must be consistently presented. Moreover, this coordination must be done inconspicuously so as not to alert subjects to the upcoming event.
Our research in this area encompasses technology for: